HS650_FinalReport_Fall2023

HW 5 Fall 2023, DSPA (HS650)

• SID: ####5432
• UMich E-mail: tashruti@umich.edu
• Declaration: I certify that the following paper represents my own independent work and conforms with the guidelines of academic honesty described in the UMich student handbook.

1. Background

I am planning to explore MIMIC data, to understand the trend and prevalence of chronic diseases in the presence of various physiological conditions amongst ICU admitted patients. Chronic Heart Failure (CHF) presents a significant challenge in healthcare, requiring a nuanced understanding of prognosis to tailor effective patient care. Typically, individuals diagnosed with CHF face a survival period of less than a year. However, recent 2023 statistics from the United States indicate a more varied picture, with certain studies suggesting an average survival of up to 5 years post-diagnosis. This discrepancy underscores the need for precise prognostic tools and individualized care plans to improve the quality of life for CHF patients.

To gain insights into the factors influencing CHF prognosis, my study delves into the Multiparameter Intelligent Monitoring in Intensive Care II (MIMIC-II) database. This publicly available resource for intensive care research enables a comprehensive analysis, incorporating variables such as age groups and SAPS scores. By exploring these dimensions, my aim is to contribute valuable information that can guide healthcare professionals in refining their prognostic assessments and, ultimately, enhancing patient outcomes.

2. Abstract

This project focuses on understanding and predicting the prognosis of Chronic Heart Failure (CHF) to improve patient care. The background research reveals a typical survival period of less than a year for CHF patients.Utilizing the Multiparameter Intelligent Monitoring in Intensive Care II (MIMIC-II) database, the study incorporates key factors such as age groups and SAPS scores to establish correlations with CHF mortality rates. The project aims to provide valuable insights into the multifaceted aspects of CHF prognosis, aiding healthcare professionals in refining their assessments and optimizing care plans. Through this, I anticipate contributing to an improved understanding of CHF and enhanced patient care strategies.

3. Preliminary Steps

Dataset This dataset was created for the purpose of a case study in the book: Secondary Analysis of Electronic Health Records, published by Springer in 2016. The dataset is derived from MIMIC-II, the publicly-accessible critical care database. The dataset (full_cohort_data.csv) is a comma-separated value file that includes a header with descriptive variable names. The main purpose of the dataset was to list the physilogical factors and intensive care parameters for a critical patient in a Hospital.

The dataset contains 1776 rows (each patient) and 46 columns (physiological and anatomical factors).

To Access the dataset Clinical data from the MIMIC-II database for a case study on indwelling arterial catheters. https://physionet.org/content/mimic2-iaccd/1.0/ The dataset in question was used throughout Chapter 16 (Data Analysis) by Raffa J. et al. to investigate the effectiveness of indwelling arterial catheters in hemodynamically stable patients with respiratory failure for mortality outcomes.

Primary Usage of the Dataset MIMIC-II provides a diverse array of patient data, including demographics, vital signs, and laboratory results, offering a comprehensive foundation for exploring various facets of intensive care research. Its extensive scope and detailed information make it a valuable tool for uncovering insights into critical care outcomes and refining healthcare practices. This dataset walks us through a meticulous examination of the impact of indwelling arterial catheters (IAC) on 28-day mortality within the intensive care unit (ICU), focusing on patients subjected to mechanical ventilation during their initial day of ICU admission.

Data Glossary SAPS score: Estimates the probability of mortality for ICU patients on admission. The SAPS III Admission Scores help sort patients based on how serious their condition is expected to be. This scoring system looks at things like the patient’s health before coming to the ICU, the reasons for their ICU admission, and how messed up their body functions are when they get to the ICU. By using this system, doctors can quickly figure out the best way to care for each patient, aiming to improve their well-being and satisfaction.

SOFA score: The Sequential Organ Failure Assessment (SOFA) score is a scoring system that assesses the performance of several organ systems in the body (neurologic, blood, liver, kidney, and blood pressure/hemodynamics).

A glimpse of datset 46 columns: Each column is a physiological factor for each patient amongst 1776 patients in the dataset.

Most of the columns seem to contain either integer or float values. Continuous data values make it easier for analysis. Descriptive Statistics:- Count: Some values are missing from various columns, as count varies in each columns Mean: age is 54, SAPS score is 14, SOFA score 5.8 Min & max: reveals the extreme values for various physiological factors, like for platelet 7 (min), 988 (max) The quantiles: if plotted will boxplot will help us determining outliers for each of the physiological factors.

Strengths and Weaknesses of the Dataset Strengths: - Reliable dataset – MIMIC is reputed and open data source for medical data - Extensive – incorporating many attributes (~46 columns) - Meaningfulness - Data dictionary is self-explanatory - Completeness – Less missing values Weakness: - Less instances – 1776 rows depicting 1776 patients - Validity – source is unknown, open-source data set and can be accessed by everyone

4. Methodology

The study would help us to have an idea on how various physiological or anatomical factors impact the prevalence of chronic diseases to overall affect the SAPS score of the patient.

PICO

P-Population The population is the patients requiring mechanical ventilation who did not require vasopressors or have a diagnosis of sepsis were identified, and the primary outcome was 28-day mortality. MIMIC data, represents population with SAPS score between ‘5-15’ identifying risk of mortality of the patient in the ICU based on the severity of the disease condition.

I-Intervention or Exposure Variable Amongst various chronic diseases, we focus on ‘Congestive heart failure (chf_flg)’ is a binary variable where 0 indicates the negative outcome and 1 indicates the positive outcome.

C-Comparison I want to look at two groups of patients: one with congestive heart failure and another without congestive heart failure. When I checked the data, I noticed that congestive heart failure and chronic renal disease show correlation of 0.25 with Mortality. The connection is stronger compared to other chronic factors. So, I decided to focus on congestive heart failure as the major factor of this study and chronic kidney disease as another variable of focus.

O-Outcome Variable “censor_flg” is the outcome variable which indicates ‘censored or death’. It is a binary variable indicative of death when equal to 0 and indicative of censored when equal to 1. Also, because the SAPS score is an indication of mortality, hence it was more relevant to choose mortality as an outcome variable.

Health Related Question**

To determine the mortality rate among individuals with SAPS scores ranging from 5 to 15, stratified by age groups, among patients admitted to the ICU, considering both those with and without heart disease.

Confounder(s) In the dataset, there are several factors that could affect the results, and one of them is a categorical variable called chronic renal disease (renal_flg). Medically, it’s known that having chronic kidney disease (CKD) makes it more likely for a person to also have heart disease [American Kidney Fund. (2022, February 15)]. CKD can lead to heart disease, and vice versa. Heart disease is actually the most common cause of death among people on dialysis. Considering renal disease as a confounding factor is important because it can impact both the chances of having heart disease and the likelihood of mortality. Another factor to consider is a continuous variable called hemoglobin count (hgb_first), which is measured when a patient is admitted to the ICU. Low hemoglobin levels in patients with congestive heart failure (CHF) have been found to be linked to a higher risk of hospitalization and overall mortality. Studies indicate a clear connection between lower hemoglobin levels and an increased risk of death. In research that treated hemoglobin as a continuous variable, even a small decrease of 1 gram per deciliter was independently associated with a significantly higher risk of mortality.

Data Preparation After initial data importing and setup, I sorted the data, cleaned it, and checked for missingness using the naniar package. I performed exploratory data analysis on various variables (both categorical and continuous) to hypothesize the question. Later, I used Clustering to find the most frequent number of clusters, PCA to check the dimensionality of the dataset, and feature selection using the boruta package before moving to the model selection and evaluating performances.

• converted numeric to binary columns

4.1 Finding Missigness

## [1] 0

• No missingness - no imputations required

4.2 2D Histogram

library(ggplot2)

ggplot(without_conf, aes(wbc_first, icu_los_day)) +
  stat_bin2d(bins=20) + 
  labs(title = "2D Histogram of WBC count vs ICU days", x = "WBC Count", y = "Number of days in ICU")  

Result Interpretation: - Through initial data exploration and visualization in our MIMIC dataset analysis, it became evident that elevated White Blood Cell (WBC) counts, associated with immunity, are linked to shorter ICU stays.

• Notably, patients with chronic diseases admitted to the ICU displayed an interesting trend. Those with an initial WBC count between 0 and 30 on the first day of ICU admission tended to have an extended stay in the ICU compared to individuals with an initial WBC count exceeding 30 on the first day of admission.

4.3 Correlational Plot

library(corrplot)
# Correlation plot
corr_conf <- cor(without_conf[1:10])

library(corrplot)
# Customizing the correlation plot
corrplot(corr_conf, type = "upper", tl.pos = "td",
         method = "circle", tl.cex = 0.5, tl.col = 'black',
         order = "hclust", diag = FALSE, mar = c(1, 1, 0, 1),
         col = colorRampPalette(c("blue", "white", "red"))(100),
         addCoef.col = "black", number.cex = 0.5,
         cl.lim = c(-1, 1), # Set the color scale limits
         title = "Correlation Plot") # Add a title

Result Interpretation: - Utilizing a correlational plot in our MIMIC dataset analysis, we aimed to identify the variable most strongly linked to the duration of ICU stays.

• The results revealed a significant correlation between the length of ICU stays and the duration of hospitalization, indicating a strong association between these two variables.

4.4 3-D Scatter Plot

Result Interpretation: - The correlational plot provided valuable insights, revealing a strong connection between the duration of ICU stays and the length of hospitalization.

• Subsequently, a 3D density plot was generated to explore the relationship between age, mortality, and the lengths of stay in both the ICU and hospital. Notably, “alive” patients are represented by green dots, while “dead” patients are denoted by blue dots.

• Findings from the plot indicate that individuals aged over 80 with chronic diseases tend to have shorter stays in both the hospital and ICU, coupled with a higher likelihood of mortality. On the other hand, patients in the 20 to 50 age group admitted to the ICU exhibit varying lengths of stay, influencing their chances of survival.

4.5 Multi-variate Bar Plot

Result Interpretation: - After categorizing the dataset into age groups below and above 60, considering the presence or absence of heart disease, our analysis unveiled higher mortality rates across the board, particularly evident in the age group above 60, especially among individuals with heart disease.

4.6 k - means Clustering

## K-means clustering with 2 clusters of sizes 629, 127
## 
## Cluster means:
##   gender_num sapsi_first   chf_flg censor_flg  renal_flg wbc_first hgb_first
## 1  0.5087440    17.82035 0.1701113  0.5755167 0.05405405  12.37587  11.93704
## 2  0.5984252    17.69291 0.1574803  0.6771654 0.04724409  12.67535  12.26693
##   icu_los_day hospital_los_day
## 1    3.153307         6.510334
## 2    9.254094        22.952756
## 
## Clustering vector:
##   [1] 1 1 1 1 2 2 1 1 1 1 2 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 1 1 1 1 1 1
##  [38] 2 2 2 1 1 1 1 1 2 2 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1
##  [75] 1 1 1 1 2 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [112] 1 1 1 1 1 1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 2 2 1 1 1 1 1 1 2
## [149] 2 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 2 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
## [186] 2 1 1 2 1 1 1 1 1 2 1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 2 2 1 1 2
## [223] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 1 2 1 1 1 1 2 1
## [260] 1 1 1 1 1 1 2 1 2 1 2 1 1 1 1 2 1 1 1 1 1 1 2 1 1 1 2 1 1 1 2 1 1 2 1 1 1
## [297] 1 1 2 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1
## [334] 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 1 1 2 1 1
## [371] 1 1 1 1 2 2 1 2 1 1 1 1 2 1 1 2 1 1 1 1 1 1 1 1 1 2 1 2 1 1 1 1 1 1 1 1 1
## [408] 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 2 2 1 1
## [445] 2 1 2 1 1 1 1 1 1 1 1 1 1 1 2 2 1 2 1 2 1 1 1 1 1 2 2 1 1 1 1 1 1 1 1 1 2
## [482] 1 1 2 1 1 1 1 1 1 1 2 1 1 1 2 1 1 1 1 1 1 1 2 1 1 1 1 1 1 2 2 2 2 1 1 2 2
## [519] 1 1 1 2 1 2 1 1 1 1 2 1 1 1 1 2 1 1 1 1 1 1 2 2 1 1 2 1 1 2 1 1 1 1 2 2 2
## [556] 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 1 1 1 1 1 1 1 1 2 1 2 1 1 1 1 1 1 1
## [593] 1 1 1 1 1 1 2 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1
## [630] 1 1 1 1 1 1 2 1 1 2 1 1 1 1 1 1 1 2 1 1 2 1 1 1 2 1 1 1 1 2 2 1 1 1 1 1 1
## [667] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 2 1 1 1 2 1 1 1 1 1 1 2 2
## [704] 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 2 1 2 2 1 2
## [741] 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 2
## 
## Within cluster sum of squares by cluster:
## [1] 54740.37 27115.97
##  (between_SS / total_SS =  28.4 %)
## 
## Available components:
## 
## [1] "cluster"      "centers"      "totss"        "withinss"     "tot.withinss"
## [6] "betweenss"    "size"         "iter"         "ifault"

## List of 1
##  $ plot.title:List of 11
##   ..$ family       : NULL
##   ..$ face         : NULL
##   ..$ colour       : NULL
##   ..$ size         : NULL
##   ..$ hjust        : num 0.5
##   ..$ vjust        : NULL
##   ..$ angle        : NULL
##   ..$ lineheight   : NULL
##   ..$ margin       : NULL
##   ..$ debug        : NULL
##   ..$ inherit.blank: logi FALSE
##   ..- attr(*, "class")= chr [1:2] "element_text" "element"
##  - attr(*, "class")= chr [1:2] "theme" "gg"
##  - attr(*, "complete")= logi FALSE
##  - attr(*, "validate")= logi TRUE

Result Interpretation: - The bar plot generated by the k-means clustering algorithm illustrates the distribution of data across 1, 2, or 3 clusters. Notably, it suggests that 2 or 3 clusters effectively capture the representation of the data.

• Further insights from the silhouette plot indicate that the optimal number of clusters representing the data is 2, reinforcing the conclusion drawn from the bar plot.

4.7 PCA

A PCA chart typically displays how the data points in a multivariate dataset project onto the principal components.

## Importance of components:
##                          PC1    PC2    PC3    PC4    PC5    PC6     PC7     PC8
## Standard deviation     1.378 1.0687 1.0545 1.0031 0.9767 0.9501 0.90159 0.80364
## Proportion of Variance 0.211 0.1269 0.1235 0.1118 0.1060 0.1003 0.09032 0.07176
## Cumulative Proportion  0.211 0.3379 0.4615 0.5733 0.6793 0.7796 0.86988 0.94165
##                            PC9
## Standard deviation     0.72470
## Proportion of Variance 0.05835
## Cumulative Proportion  1.00000

library(ggbiplot)
# Basic ggbiplot
ggbiplot(without_conf.pca, obs.scale = 1, var.scale = 1, groups = without_conf$GroupingVariable)

# Customized ggbiplot with additional parameters
ggbiplot(without_conf.pca,
         obs.scale = 1, var.scale = 1,
         groups = without_conf$GroupingVariable,
         ellipse = TRUE, ellipse.prob = 0.68, # Add ellipses with 68% confidence
         circle = TRUE, # Add a unit circle
         varname.size = 4, varname.adjust = 1.5, # Adjust size and position of variable names
         varname.abbrev = TRUE, # Abbreviate variable names
         var.axes = TRUE, # Add arrows for variable axes
         labels = rownames(without_conf), # Add labels for data points
         label.size = 3, # Adjust size of labels
         alpha = 0.5, # Set transparency of points
         shape = without_conf$ShapeVariable, # Use different shapes for data points
         legend = TRUE, # Display legend
         biplot = TRUE, # Display biplot
         loadings = "variance", # Show variable loadings as proportion of variance
         loadings.label = TRUE, # Show variable names for loadings
         loadings.label.size = 3, # Adjust size of loading labels
         varname.col = "blue", # Set color for variable names
         varname.face = "bold", # Set font weight for variable names
         varname.label = TRUE, # Show variable labels
         varname.label.size = 3, # Adjust size of variable labels
         varname.label.col = "red" # Set color for variable labels
)

Result Interpretation: - The score plots reveal how the data is projected along the principal components. Points located farther from the center are likely outliers or represent naturally extreme observations.

• Notably, a majority of data points exhibit a first principal component (PC) score of 0, while a few data points deviate with a first PC score of -2.

4.8 Feature Selection

## Boruta performed 99 iterations in 1.649661 mins.
##  8 attributes confirmed important: age, agegroup, censor_flg,
## hgb_first, icu_los_day and 3 more;
##  1 attributes confirmed unimportant: gender_num;
##  3 tentative attributes left: chf_flg, heart_failure, hospital_los_day;

Now selected sapsi_first, heart_failure, mortality, using tentative and confirmed important attributes:

##           age gender_num  chf_flg censor_flg renal_flg wbc_first hgb_first
## [1,] 12.12949 -0.9084812 1.762399   6.906130  4.389413  4.161883  7.152791
## [2,] 13.14845  2.2868219 2.122811   8.523155  4.142409  3.319650  7.362015
## [3,] 11.18096  0.6705264 3.687560   4.390831  4.074064  3.197270  5.621322
## [4,] 12.27479  0.8830721 2.278991   7.654170  5.412227  3.430496  7.397266
## [5,] 12.29947  0.2566831 1.720871   6.827953  7.079582  3.454351  6.018177
## [6,] 11.96974 -0.6589906 2.050596   5.877351  4.497742  2.564088  8.042439
##      icu_los_day hospital_los_day agegroup
## [1,]    5.499151        4.8730066 6.538453
## [2,]    3.897263        1.5475419 6.804577
## [3,]    5.478581        0.9876445 6.977166
## [4,]    6.337409        3.7439084 6.493189
## [5,]    4.225862        1.1968753 8.459238
## [6,]    6.070956        0.1418116 8.267406

## Boruta performed 99 iterations in 1.649661 mins.
## Tentatives roughfixed over the last 99 iterations.
##  10 attributes confirmed important: age, agegroup, censor_flg, chf_flg,
## heart_failure and 5 more;
##  2 attributes confirmed unimportant: gender_num, hospital_los_day;

Result Interpretation: - After removing features with infinite importance, we organized the remaining significant features by their median importance. Using plotly, we illustrated these features through boxplots, showcasing their median, quartiles, and minimum and maximum values.

• The whiskerplot enables a clear view of the importance score distribution for each variable, aiding in the identification of tentative and crucial variables. This visualization assists in assessing the range of importance scores within a single variable and may guide decisions on whether to retain or discard tentative features.

4.9 Modeling and model evaluation

4.9.1 Normalizing Data

• Checking if normalization worked

summary(without_conf_n$sapsi_first)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
## 0.00000 0.05882 0.11765 0.16464 0.23529 1.00000

4.9.2 Partition data

4.9.3 Rpart

## n= 572 
## 
## node), split, n, loss, yval, (yprob)
##       * denotes terminal node
## 
##   1) root 572 224 1 (0.3916084 0.6083916)  
##     2) age>=0.5790757 325 142 0 (0.5630769 0.4369231)  
##       4) age>=0.8433225 64  15 0 (0.7656250 0.2343750)  
##         8) icu_los_day>=0.01825741 57  10 0 (0.8245614 0.1754386) *
##         9) icu_los_day< 0.01825741 7   2 1 (0.2857143 0.7142857) *
##       5) age< 0.8433225 261 127 0 (0.5134100 0.4865900)  
##        10) age>=0.702679 154  66 0 (0.5714286 0.4285714)  
##          20) icu_los_day>=0.3714751 7   0 0 (1.0000000 0.0000000) *
##          21) icu_los_day< 0.3714751 147  66 0 (0.5510204 0.4489796)  
##            42) icu_los_day< 0.06381056 62  21 0 (0.6612903 0.3387097) *
##            43) icu_los_day>=0.06381056 85  40 1 (0.4705882 0.5294118)  
##              86) icu_los_day>=0.1930586 22   7 0 (0.6818182 0.3181818) *
##              87) icu_los_day< 0.1930586 63  25 1 (0.3968254 0.6031746)  
##               174) age< 0.7783392 32  15 1 (0.4687500 0.5312500)  
##                 348) age>=0.751735 12   3 0 (0.7500000 0.2500000) *
##                 349) age< 0.751735 20   6 1 (0.3000000 0.7000000) *
##               175) age>=0.7783392 31  10 1 (0.3225806 0.6774194) *
##        11) age< 0.702679 107  46 1 (0.4299065 0.5700935)  
##          22) age< 0.6503991 52  24 0 (0.5384615 0.4615385)  
##            44) age>=0.642669 9   0 0 (1.0000000 0.0000000) *
##            45) age< 0.642669 43  19 1 (0.4418605 0.5581395) *
##          23) age>=0.6503991 55  18 1 (0.3272727 0.6727273) *
##     3) age< 0.5790757 247  41 1 (0.1659919 0.8340081) *

Result Interpretation: - We divided the dataset into 75% training data and 25% test data, employing the rpart algorithm to build a classification tree. The plot above illustrates the crucial features utilized by the algorithm for observation classification.

• Notably, Age and icu_los_day stood out as the most important variables for the recursive partitioning process. The plot indicates that for individuals aged above 58, a predominant outcome is “0,” implying a higher incidence of patient deaths in the ICU for those aged 58 or older.

4.9.4 K Nearest Neighbours

## [1] "Accuracy = 0.374338624338624"

Interpretation & Analysis: - Experimenting with different values of k to train the model revealed that the highest accuracy achieved was 38% for k=28, indicating a relatively low accuracy level. In response, we proceeded to train the dataset using the Logistic Regression algorithm.

4.9.5 Linear Models (Logistic Regression)

##              age       gender_num      sapsi_first          chf_flg 
##       0.23401548       0.49976343       0.15878730       0.37410439 
##       censor_flg        renal_flg        wbc_first        hgb_first 
##       0.49167711       0.22400212       0.06767688       0.13801630 
##      icu_los_day hospital_los_day 
##       0.13859862       0.08316255

## 
## Call:
## glm(formula = censor_flg ~ age + chf_flg + sapsi_first + icu_los_day, 
##     family = "binomial", data = training)
## 
## Deviance Residuals: 
##     Min       1Q   Median       3Q      Max  
## -2.4671  -0.9968   0.4035   0.9178   1.8318  
## 
## Coefficients:
##             Estimate Std. Error z value Pr(>|z|)    
## (Intercept)  3.80406    0.39175   9.710  < 2e-16 ***
## age         -4.73250    0.55722  -8.493  < 2e-16 ***
## chf_flg     -0.06498    0.25421  -0.256  0.79826    
## sapsi_first -1.64142    0.60815  -2.699  0.00695 ** 
## icu_los_day -1.11898    0.73715  -1.518  0.12902    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 1)
## 
##     Null deviance: 765.86  on 571  degrees of freedom
## Residual deviance: 634.26  on 567  degrees of freedom
## AIC: 644.26
## 
## Number of Fisher Scoring iterations: 4

#calculate probability of default for each individual in test dataset
predicted <- predict(lr, testing, type="response")

#calculate AUC
library(pROC)
auc(testing$censor_flg, predicted)

## Area under the curve: 0.7146

Result Interpretation: - A greater Area Under the Curve (AUC) signifies enhanced model performance, indicating its proficiency in distinguishing between positive and negative classes. With an AUC of 0.71, this model’s predictions demonstrate moderate accuracy and leave room for improvement.

• The Akaike Information Criterion (AIC) stands at 644.26. A smaller AIC value indicates a better fit for the model, emphasizing the quality of its overall performance.

5. Results

I experimented with two models – KNN and logistic regression. The logistic regression model emerged as the most suitable fit, with an Area Under the Curve (AUC) of 0.71. This suggests that the model achieved a 71% accuracy rate in predicting mortality, taking into account factors such as age, SAPS score at ICU admission, and the presence or absence of congestive heart failure (chf_flg=1 or chf_flg=0).

6. Discussion

For associations and correlations to be scientifically valid, it’s crucial to ensure their reliability. In future analyses of this inquiry, we can enhance the model’s robustness by collecting additional data and devising strategies to select features that truly represent the sample, making the results more meaningful. Following internal validation of the model, it’s advisable to conduct pilot tests in different geographic areas for external validation, addressing any discrepancies before implementing it in real-world scenarios. It’s imperative to disclose regulations preventing the model’s misuse by for-profit agencies, particularly in adjusting insurance premiums based on health conditions, to prevent potential disparities.

7. References

[1] American Heart Association. (2022). Kidney Disease & Cardiovascular Disease. Website: https://www.heart.org/en/health-topics/kidney-disease/kidney-disease-and-heart-disease

[2] North Carolina Institute for Public Health. (2015). Chronic Kidney Disease in North Carolina: A Public Health Surveillance Report. Report: https://sph.unc.edu/wp-content/uploads/sites/112/2015/07/nciph_ERIC11.pdf

[3] Hill, A. B. (1965). The Environment and Disease: Association or Causation? Paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2126329/

[4] Go, A. S., et al. (2015). Heart Disease and Stroke Statistics—2015 Update. Article: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4589117/

[5] World Health Organization. (2023). Cardiovascular Diseases (CVDs). Fact Sheet: https://www.who.int/news-room/fact-sheets/detail/cardiovascular-diseases-(cvds)

[6] Lim, W. Y., et al. (2013). Ischemic Heart Disease (IHD) Mortality among Chinese and Malays in Singapore: Ethnic Variations Uncovered. Article: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3760015/

[7] Morelli, A., et al. (2005). Effects of Levosimendan on Systemic and Regional Hemodynamics in Septic Myocardial Depression. Article: https://link.springer.com/article/10.1007/s00134-005-2763-5

[8] Mishra, J., et al. (2021). Association Between Anemia and Mortality in Patients With Chronic Heart Failure: A Systematic Review and Meta-Analysis. Article: https://www.nature.com/articles/s41598-021-03397-3.pdf?proof=t+target%3D

[9] Díez-Delhoyo, F., et al. (2021). Heart Failure in Older Adults: A Comprehensive Review from the International Heart Failure Association.Article: https://www.frontiersin.org/articles/10.3389/fcvm.2021.774935/full

[10] Haq, A., et al. (2014). The Simplified Acute Physiology Score III Is Superior to the Simplified Acute Physiology Score II and Acute Physiology and Chronic Health Evaluation II in Predicting Surgical and ICU Mortality in the “Oldest Old”. Article: https://www.hindawi.com/journals/cggr/2014/934852/

[11] Marik, P. E. (2006). Management of the Critically Ill Geriatric Patient. Article: https://journals.lww.com/ccmjournal/Citation/2006/09001/Management_of_the_critically_ill_geriatric_patient.20.aspx

[12] Tang, Y. D., & Katz, S. D. (2006). Anemia in Chronic Heart Failure: Prevalence, Etiology, Clinical Correlates, and Treatment Options.Article: https://www.ahajournals.org/doi/full/10.1161/CIRCULATIONAHA.105.588986

[13] Samaritan Healthcare & Hospice. (2020). End-Stage Heart Failure: What to Expect.Website: https://samaritannj.org/hospice-blog-and-events/hospice-palliative-care-blog/end-stage-heart-failure-what-to-expect/